Bigtable's SQL support is now in GA and supports server-side aggregations. In addition to read-time aggregations it is also possible to define incremental materialized views that do rollups/aggregations at ingest-time.

In a lot of real-time analytics applications, one can imagine stacking these e.g. use an incremental materialized view to aggregate clicks into 15 minute windows and then at read time apply filters and GROUP BY to convert pre-aggregated data into a coarser granularity like N hours, days etc.